1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
3 #include <linux/errno.h>
4 #include <linux/fs.h>
5 #include <linux/file.h>
6 #include <linux/mm.h>
7 #include <linux/slab.h>
8 #include <linux/poll.h>
9 #include <linux/hashtable.h>
10 #include <linux/io_uring.h>
11
12 #include <trace/events/io_uring.h>
13
14 #include <uapi/linux/io_uring.h>
15
16 #include "io_uring.h"
17 #include "alloc_cache.h"
18 #include "refs.h"
19 #include "napi.h"
20 #include "opdef.h"
21 #include "kbuf.h"
22 #include "poll.h"
23 #include "cancel.h"
24
25 struct io_poll_update {
26 struct file *file;
27 u64 old_user_data;
28 u64 new_user_data;
29 __poll_t events;
30 bool update_events;
31 bool update_user_data;
32 };
33
34 struct io_poll_table {
35 struct poll_table_struct pt;
36 struct io_kiocb *req;
37 int nr_entries;
38 int error;
39 bool owning;
40 /* output value, set only if arm poll returns >0 */
41 __poll_t result_mask;
42 };
43
44 #define IO_POLL_CANCEL_FLAG BIT(31)
45 #define IO_POLL_RETRY_FLAG BIT(30)
46 #define IO_POLL_REF_MASK GENMASK(29, 0)
47
48 /*
49 * We usually have 1-2 refs taken, 128 is more than enough and we want to
50 * maximise the margin between this amount and the moment when it overflows.
51 */
52 #define IO_POLL_REF_BIAS 128
53
54 #define IO_WQE_F_DOUBLE 1
55
56 static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
57 void *key);
58
wqe_to_req(struct wait_queue_entry * wqe)59 static inline struct io_kiocb *wqe_to_req(struct wait_queue_entry *wqe)
60 {
61 unsigned long priv = (unsigned long)wqe->private;
62
63 return (struct io_kiocb *)(priv & ~IO_WQE_F_DOUBLE);
64 }
65
wqe_is_double(struct wait_queue_entry * wqe)66 static inline bool wqe_is_double(struct wait_queue_entry *wqe)
67 {
68 unsigned long priv = (unsigned long)wqe->private;
69
70 return priv & IO_WQE_F_DOUBLE;
71 }
72
io_poll_get_ownership_slowpath(struct io_kiocb * req)73 static bool io_poll_get_ownership_slowpath(struct io_kiocb *req)
74 {
75 int v;
76
77 /*
78 * poll_refs are already elevated and we don't have much hope for
79 * grabbing the ownership. Instead of incrementing set a retry flag
80 * to notify the loop that there might have been some change.
81 */
82 v = atomic_fetch_or(IO_POLL_RETRY_FLAG, &req->poll_refs);
83 if (v & IO_POLL_REF_MASK)
84 return false;
85 return !(atomic_fetch_inc(&req->poll_refs) & IO_POLL_REF_MASK);
86 }
87
88 /*
89 * If refs part of ->poll_refs (see IO_POLL_REF_MASK) is 0, it's free. We can
90 * bump it and acquire ownership. It's disallowed to modify requests while not
91 * owning it, that prevents from races for enqueueing task_work's and b/w
92 * arming poll and wakeups.
93 */
io_poll_get_ownership(struct io_kiocb * req)94 static inline bool io_poll_get_ownership(struct io_kiocb *req)
95 {
96 if (unlikely(atomic_read(&req->poll_refs) >= IO_POLL_REF_BIAS))
97 return io_poll_get_ownership_slowpath(req);
98 return !(atomic_fetch_inc(&req->poll_refs) & IO_POLL_REF_MASK);
99 }
100
io_poll_mark_cancelled(struct io_kiocb * req)101 static void io_poll_mark_cancelled(struct io_kiocb *req)
102 {
103 atomic_or(IO_POLL_CANCEL_FLAG, &req->poll_refs);
104 }
105
io_poll_get_double(struct io_kiocb * req)106 static struct io_poll *io_poll_get_double(struct io_kiocb *req)
107 {
108 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
109 if (req->opcode == IORING_OP_POLL_ADD)
110 return req->async_data;
111 return req->apoll->double_poll;
112 }
113
io_poll_get_single(struct io_kiocb * req)114 static struct io_poll *io_poll_get_single(struct io_kiocb *req)
115 {
116 if (req->opcode == IORING_OP_POLL_ADD)
117 return io_kiocb_to_cmd(req, struct io_poll);
118 return &req->apoll->poll;
119 }
120
io_poll_req_insert(struct io_kiocb * req)121 static void io_poll_req_insert(struct io_kiocb *req)
122 {
123 struct io_hash_table *table = &req->ctx->cancel_table;
124 u32 index = hash_long(req->cqe.user_data, table->hash_bits);
125
126 lockdep_assert_held(&req->ctx->uring_lock);
127
128 hlist_add_head(&req->hash_node, &table->hbs[index].list);
129 }
130
io_init_poll_iocb(struct io_poll * poll,__poll_t events)131 static void io_init_poll_iocb(struct io_poll *poll, __poll_t events)
132 {
133 poll->head = NULL;
134 #define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP)
135 /* mask in events that we always want/need */
136 poll->events = events | IO_POLL_UNMASK;
137 INIT_LIST_HEAD(&poll->wait.entry);
138 init_waitqueue_func_entry(&poll->wait, io_poll_wake);
139 }
140
io_poll_remove_entry(struct io_poll * poll)141 static inline void io_poll_remove_entry(struct io_poll *poll)
142 {
143 struct wait_queue_head *head = smp_load_acquire(&poll->head);
144
145 if (head) {
146 spin_lock_irq(&head->lock);
147 list_del_init(&poll->wait.entry);
148 poll->head = NULL;
149 spin_unlock_irq(&head->lock);
150 }
151 }
152
io_poll_remove_entries(struct io_kiocb * req)153 static void io_poll_remove_entries(struct io_kiocb *req)
154 {
155 /*
156 * Nothing to do if neither of those flags are set. Avoid dipping
157 * into the poll/apoll/double cachelines if we can.
158 */
159 if (!(req->flags & (REQ_F_SINGLE_POLL | REQ_F_DOUBLE_POLL)))
160 return;
161
162 /*
163 * While we hold the waitqueue lock and the waitqueue is nonempty,
164 * wake_up_pollfree() will wait for us. However, taking the waitqueue
165 * lock in the first place can race with the waitqueue being freed.
166 *
167 * We solve this as eventpoll does: by taking advantage of the fact that
168 * all users of wake_up_pollfree() will RCU-delay the actual free. If
169 * we enter rcu_read_lock() and see that the pointer to the queue is
170 * non-NULL, we can then lock it without the memory being freed out from
171 * under us.
172 *
173 * Keep holding rcu_read_lock() as long as we hold the queue lock, in
174 * case the caller deletes the entry from the queue, leaving it empty.
175 * In that case, only RCU prevents the queue memory from being freed.
176 */
177 rcu_read_lock();
178 if (req->flags & REQ_F_SINGLE_POLL)
179 io_poll_remove_entry(io_poll_get_single(req));
180 if (req->flags & REQ_F_DOUBLE_POLL)
181 io_poll_remove_entry(io_poll_get_double(req));
182 rcu_read_unlock();
183 }
184
185 enum {
186 IOU_POLL_DONE = 0,
187 IOU_POLL_NO_ACTION = 1,
188 IOU_POLL_REMOVE_POLL_USE_RES = 2,
189 IOU_POLL_REISSUE = 3,
190 IOU_POLL_REQUEUE = 4,
191 };
192
__io_poll_execute(struct io_kiocb * req,int mask)193 static void __io_poll_execute(struct io_kiocb *req, int mask)
194 {
195 unsigned flags = 0;
196
197 io_req_set_res(req, mask, 0);
198 req->io_task_work.func = io_poll_task_func;
199
200 trace_io_uring_task_add(req, mask);
201
202 if (!(req->flags & REQ_F_POLL_NO_LAZY))
203 flags = IOU_F_TWQ_LAZY_WAKE;
204 __io_req_task_work_add(req, flags);
205 }
206
io_poll_execute(struct io_kiocb * req,int res)207 static inline void io_poll_execute(struct io_kiocb *req, int res)
208 {
209 if (io_poll_get_ownership(req))
210 __io_poll_execute(req, res);
211 }
212
213 /*
214 * All poll tw should go through this. Checks for poll events, manages
215 * references, does rewait, etc.
216 *
217 * Returns a negative error on failure. IOU_POLL_NO_ACTION when no action
218 * require, which is either spurious wakeup or multishot CQE is served.
219 * IOU_POLL_DONE when it's done with the request, then the mask is stored in
220 * req->cqe.res. IOU_POLL_REMOVE_POLL_USE_RES indicates to remove multishot
221 * poll and that the result is stored in req->cqe.
222 */
io_poll_check_events(struct io_kiocb * req,struct io_tw_state * ts)223 static int io_poll_check_events(struct io_kiocb *req, struct io_tw_state *ts)
224 {
225 int v;
226
227 if (unlikely(io_should_terminate_tw()))
228 return -ECANCELED;
229
230 do {
231 v = atomic_read(&req->poll_refs);
232
233 if (unlikely(v != 1)) {
234 /* tw should be the owner and so have some refs */
235 if (WARN_ON_ONCE(!(v & IO_POLL_REF_MASK)))
236 return IOU_POLL_NO_ACTION;
237 if (v & IO_POLL_CANCEL_FLAG)
238 return -ECANCELED;
239 /*
240 * cqe.res contains only events of the first wake up
241 * and all others are to be lost. Redo vfs_poll() to get
242 * up to date state.
243 */
244 if ((v & IO_POLL_REF_MASK) != 1)
245 req->cqe.res = 0;
246
247 if (v & IO_POLL_RETRY_FLAG) {
248 req->cqe.res = 0;
249 /*
250 * We won't find new events that came in between
251 * vfs_poll and the ref put unless we clear the
252 * flag in advance.
253 */
254 atomic_andnot(IO_POLL_RETRY_FLAG, &req->poll_refs);
255 v &= ~IO_POLL_RETRY_FLAG;
256 }
257 }
258
259 /* the mask was stashed in __io_poll_execute */
260 if (!req->cqe.res) {
261 struct poll_table_struct pt = { ._key = req->apoll_events };
262 req->cqe.res = vfs_poll(req->file, &pt) & req->apoll_events;
263 /*
264 * We got woken with a mask, but someone else got to
265 * it first. The above vfs_poll() doesn't add us back
266 * to the waitqueue, so if we get nothing back, we
267 * should be safe and attempt a reissue.
268 */
269 if (unlikely(!req->cqe.res)) {
270 /* Multishot armed need not reissue */
271 if (!(req->apoll_events & EPOLLONESHOT))
272 continue;
273 return IOU_POLL_REISSUE;
274 }
275 }
276 if (req->apoll_events & EPOLLONESHOT)
277 return IOU_POLL_DONE;
278
279 /* multishot, just fill a CQE and proceed */
280 if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
281 __poll_t mask = mangle_poll(req->cqe.res &
282 req->apoll_events);
283
284 if (!io_req_post_cqe(req, mask, IORING_CQE_F_MORE)) {
285 io_req_set_res(req, mask, 0);
286 return IOU_POLL_REMOVE_POLL_USE_RES;
287 }
288 } else {
289 int ret = io_poll_issue(req, ts);
290 if (ret == IOU_STOP_MULTISHOT)
291 return IOU_POLL_REMOVE_POLL_USE_RES;
292 else if (ret == IOU_REQUEUE)
293 return IOU_POLL_REQUEUE;
294 if (ret < 0)
295 return ret;
296 }
297
298 /* force the next iteration to vfs_poll() */
299 req->cqe.res = 0;
300
301 /*
302 * Release all references, retry if someone tried to restart
303 * task_work while we were executing it.
304 */
305 v &= IO_POLL_REF_MASK;
306 } while (atomic_sub_return(v, &req->poll_refs) & IO_POLL_REF_MASK);
307
308 io_napi_add(req);
309 return IOU_POLL_NO_ACTION;
310 }
311
io_poll_task_func(struct io_kiocb * req,struct io_tw_state * ts)312 void io_poll_task_func(struct io_kiocb *req, struct io_tw_state *ts)
313 {
314 int ret;
315
316 ret = io_poll_check_events(req, ts);
317 if (ret == IOU_POLL_NO_ACTION) {
318 return;
319 } else if (ret == IOU_POLL_REQUEUE) {
320 __io_poll_execute(req, 0);
321 return;
322 }
323 io_poll_remove_entries(req);
324 /* task_work always has ->uring_lock held */
325 hash_del(&req->hash_node);
326
327 if (req->opcode == IORING_OP_POLL_ADD) {
328 if (ret == IOU_POLL_DONE) {
329 struct io_poll *poll;
330
331 poll = io_kiocb_to_cmd(req, struct io_poll);
332 req->cqe.res = mangle_poll(req->cqe.res & poll->events);
333 } else if (ret == IOU_POLL_REISSUE) {
334 io_req_task_submit(req, ts);
335 return;
336 } else if (ret != IOU_POLL_REMOVE_POLL_USE_RES) {
337 req->cqe.res = ret;
338 req_set_fail(req);
339 }
340
341 io_req_set_res(req, req->cqe.res, 0);
342 io_req_task_complete(req, ts);
343 } else {
344 io_tw_lock(req->ctx, ts);
345
346 if (ret == IOU_POLL_REMOVE_POLL_USE_RES)
347 io_req_task_complete(req, ts);
348 else if (ret == IOU_POLL_DONE || ret == IOU_POLL_REISSUE)
349 io_req_task_submit(req, ts);
350 else
351 io_req_defer_failed(req, ret);
352 }
353 }
354
io_poll_cancel_req(struct io_kiocb * req)355 static void io_poll_cancel_req(struct io_kiocb *req)
356 {
357 io_poll_mark_cancelled(req);
358 /* kick tw, which should complete the request */
359 io_poll_execute(req, 0);
360 }
361
362 #define IO_ASYNC_POLL_COMMON (EPOLLONESHOT | EPOLLPRI)
363
io_pollfree_wake(struct io_kiocb * req,struct io_poll * poll)364 static __cold int io_pollfree_wake(struct io_kiocb *req, struct io_poll *poll)
365 {
366 io_poll_mark_cancelled(req);
367 /* we have to kick tw in case it's not already */
368 io_poll_execute(req, 0);
369
370 /*
371 * If the waitqueue is being freed early but someone is already
372 * holds ownership over it, we have to tear down the request as
373 * best we can. That means immediately removing the request from
374 * its waitqueue and preventing all further accesses to the
375 * waitqueue via the request.
376 */
377 list_del_init(&poll->wait.entry);
378
379 /*
380 * Careful: this *must* be the last step, since as soon
381 * as req->head is NULL'ed out, the request can be
382 * completed and freed, since aio_poll_complete_work()
383 * will no longer need to take the waitqueue lock.
384 */
385 smp_store_release(&poll->head, NULL);
386 return 1;
387 }
388
io_poll_wake(struct wait_queue_entry * wait,unsigned mode,int sync,void * key)389 static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
390 void *key)
391 {
392 struct io_kiocb *req = wqe_to_req(wait);
393 struct io_poll *poll = container_of(wait, struct io_poll, wait);
394 __poll_t mask = key_to_poll(key);
395
396 if (unlikely(mask & POLLFREE))
397 return io_pollfree_wake(req, poll);
398
399 /* for instances that support it check for an event match first */
400 if (mask && !(mask & (poll->events & ~IO_ASYNC_POLL_COMMON)))
401 return 0;
402
403 if (io_poll_get_ownership(req)) {
404 /*
405 * If we trigger a multishot poll off our own wakeup path,
406 * disable multishot as there is a circular dependency between
407 * CQ posting and triggering the event.
408 */
409 if (mask & EPOLL_URING_WAKE)
410 poll->events |= EPOLLONESHOT;
411
412 /* optional, saves extra locking for removal in tw handler */
413 if (mask && poll->events & EPOLLONESHOT) {
414 list_del_init(&poll->wait.entry);
415 poll->head = NULL;
416 if (wqe_is_double(wait))
417 req->flags &= ~REQ_F_DOUBLE_POLL;
418 else
419 req->flags &= ~REQ_F_SINGLE_POLL;
420 }
421 __io_poll_execute(req, mask);
422 }
423 return 1;
424 }
425
426 /* fails only when polling is already completing by the first entry */
io_poll_double_prepare(struct io_kiocb * req)427 static bool io_poll_double_prepare(struct io_kiocb *req)
428 {
429 struct wait_queue_head *head;
430 struct io_poll *poll = io_poll_get_single(req);
431
432 /* head is RCU protected, see io_poll_remove_entries() comments */
433 rcu_read_lock();
434 head = smp_load_acquire(&poll->head);
435 /*
436 * poll arm might not hold ownership and so race for req->flags with
437 * io_poll_wake(). There is only one poll entry queued, serialise with
438 * it by taking its head lock. As we're still arming the tw hanlder
439 * is not going to be run, so there are no races with it.
440 */
441 if (head) {
442 spin_lock_irq(&head->lock);
443 req->flags |= REQ_F_DOUBLE_POLL;
444 if (req->opcode == IORING_OP_POLL_ADD)
445 req->flags |= REQ_F_ASYNC_DATA;
446 spin_unlock_irq(&head->lock);
447 }
448 rcu_read_unlock();
449 return !!head;
450 }
451
__io_queue_proc(struct io_poll * poll,struct io_poll_table * pt,struct wait_queue_head * head,struct io_poll ** poll_ptr)452 static void __io_queue_proc(struct io_poll *poll, struct io_poll_table *pt,
453 struct wait_queue_head *head,
454 struct io_poll **poll_ptr)
455 {
456 struct io_kiocb *req = pt->req;
457 unsigned long wqe_private = (unsigned long) req;
458
459 /*
460 * The file being polled uses multiple waitqueues for poll handling
461 * (e.g. one for read, one for write). Setup a separate io_poll
462 * if this happens.
463 */
464 if (unlikely(pt->nr_entries)) {
465 struct io_poll *first = poll;
466
467 /* double add on the same waitqueue head, ignore */
468 if (first->head == head)
469 return;
470 /* already have a 2nd entry, fail a third attempt */
471 if (*poll_ptr) {
472 if ((*poll_ptr)->head == head)
473 return;
474 pt->error = -EINVAL;
475 return;
476 }
477
478 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
479 if (!poll) {
480 pt->error = -ENOMEM;
481 return;
482 }
483
484 /* mark as double wq entry */
485 wqe_private |= IO_WQE_F_DOUBLE;
486 io_init_poll_iocb(poll, first->events);
487 if (!io_poll_double_prepare(req)) {
488 /* the request is completing, just back off */
489 kfree(poll);
490 return;
491 }
492 *poll_ptr = poll;
493 } else {
494 /* fine to modify, there is no poll queued to race with us */
495 req->flags |= REQ_F_SINGLE_POLL;
496 }
497
498 pt->nr_entries++;
499 poll->head = head;
500 poll->wait.private = (void *) wqe_private;
501
502 if (poll->events & EPOLLEXCLUSIVE) {
503 add_wait_queue_exclusive(head, &poll->wait);
504 } else {
505 add_wait_queue(head, &poll->wait);
506 }
507 }
508
io_poll_queue_proc(struct file * file,struct wait_queue_head * head,struct poll_table_struct * p)509 static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
510 struct poll_table_struct *p)
511 {
512 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
513 struct io_poll *poll = io_kiocb_to_cmd(pt->req, struct io_poll);
514
515 __io_queue_proc(poll, pt, head,
516 (struct io_poll **) &pt->req->async_data);
517 }
518
io_poll_can_finish_inline(struct io_kiocb * req,struct io_poll_table * pt)519 static bool io_poll_can_finish_inline(struct io_kiocb *req,
520 struct io_poll_table *pt)
521 {
522 return pt->owning || io_poll_get_ownership(req);
523 }
524
io_poll_add_hash(struct io_kiocb * req,unsigned int issue_flags)525 static void io_poll_add_hash(struct io_kiocb *req, unsigned int issue_flags)
526 {
527 struct io_ring_ctx *ctx = req->ctx;
528
529 io_ring_submit_lock(ctx, issue_flags);
530 io_poll_req_insert(req);
531 io_ring_submit_unlock(ctx, issue_flags);
532 }
533
534 /*
535 * Returns 0 when it's handed over for polling. The caller owns the requests if
536 * it returns non-zero, but otherwise should not touch it. Negative values
537 * contain an error code. When the result is >0, the polling has completed
538 * inline and ipt.result_mask is set to the mask.
539 */
__io_arm_poll_handler(struct io_kiocb * req,struct io_poll * poll,struct io_poll_table * ipt,__poll_t mask,unsigned issue_flags)540 static int __io_arm_poll_handler(struct io_kiocb *req,
541 struct io_poll *poll,
542 struct io_poll_table *ipt, __poll_t mask,
543 unsigned issue_flags)
544 {
545 INIT_HLIST_NODE(&req->hash_node);
546 io_init_poll_iocb(poll, mask);
547 poll->file = req->file;
548 req->apoll_events = poll->events;
549
550 ipt->pt._key = mask;
551 ipt->req = req;
552 ipt->error = 0;
553 ipt->nr_entries = 0;
554 /*
555 * Polling is either completed here or via task_work, so if we're in the
556 * task context we're naturally serialised with tw by merit of running
557 * the same task. When it's io-wq, take the ownership to prevent tw
558 * from running. However, when we're in the task context, skip taking
559 * it as an optimisation.
560 *
561 * Note: even though the request won't be completed/freed, without
562 * ownership we still can race with io_poll_wake().
563 * io_poll_can_finish_inline() tries to deal with that.
564 */
565 ipt->owning = issue_flags & IO_URING_F_UNLOCKED;
566 atomic_set(&req->poll_refs, (int)ipt->owning);
567
568 /*
569 * Exclusive waits may only wake a limited amount of entries
570 * rather than all of them, this may interfere with lazy
571 * wake if someone does wait(events > 1). Ensure we don't do
572 * lazy wake for those, as we need to process each one as they
573 * come in.
574 */
575 if (poll->events & EPOLLEXCLUSIVE)
576 req->flags |= REQ_F_POLL_NO_LAZY;
577
578 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
579
580 if (unlikely(ipt->error || !ipt->nr_entries)) {
581 io_poll_remove_entries(req);
582
583 if (!io_poll_can_finish_inline(req, ipt)) {
584 io_poll_mark_cancelled(req);
585 return 0;
586 } else if (mask && (poll->events & EPOLLET)) {
587 ipt->result_mask = mask;
588 return 1;
589 }
590 return ipt->error ?: -EINVAL;
591 }
592
593 if (mask &&
594 ((poll->events & (EPOLLET|EPOLLONESHOT)) == (EPOLLET|EPOLLONESHOT))) {
595 if (!io_poll_can_finish_inline(req, ipt)) {
596 io_poll_add_hash(req, issue_flags);
597 return 0;
598 }
599 io_poll_remove_entries(req);
600 ipt->result_mask = mask;
601 /* no one else has access to the req, forget about the ref */
602 return 1;
603 }
604
605 io_poll_add_hash(req, issue_flags);
606
607 if (mask && (poll->events & EPOLLET) &&
608 io_poll_can_finish_inline(req, ipt)) {
609 __io_poll_execute(req, mask);
610 return 0;
611 }
612 io_napi_add(req);
613
614 if (ipt->owning) {
615 /*
616 * Try to release ownership. If we see a change of state, e.g.
617 * poll was waken up, queue up a tw, it'll deal with it.
618 */
619 if (atomic_cmpxchg(&req->poll_refs, 1, 0) != 1)
620 __io_poll_execute(req, 0);
621 }
622 return 0;
623 }
624
io_async_queue_proc(struct file * file,struct wait_queue_head * head,struct poll_table_struct * p)625 static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
626 struct poll_table_struct *p)
627 {
628 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
629 struct async_poll *apoll = pt->req->apoll;
630
631 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
632 }
633
634 /*
635 * We can't reliably detect loops in repeated poll triggers and issue
636 * subsequently failing. But rather than fail these immediately, allow a
637 * certain amount of retries before we give up. Given that this condition
638 * should _rarely_ trigger even once, we should be fine with a larger value.
639 */
640 #define APOLL_MAX_RETRY 128
641
io_req_alloc_apoll(struct io_kiocb * req,unsigned issue_flags)642 static struct async_poll *io_req_alloc_apoll(struct io_kiocb *req,
643 unsigned issue_flags)
644 {
645 struct io_ring_ctx *ctx = req->ctx;
646 struct async_poll *apoll;
647
648 if (req->flags & REQ_F_POLLED) {
649 apoll = req->apoll;
650 kfree(apoll->double_poll);
651 } else if (!(issue_flags & IO_URING_F_UNLOCKED)) {
652 apoll = io_alloc_cache_get(&ctx->apoll_cache);
653 if (!apoll)
654 goto alloc_apoll;
655 apoll->poll.retries = APOLL_MAX_RETRY;
656 } else {
657 alloc_apoll:
658 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
659 if (unlikely(!apoll))
660 return NULL;
661 apoll->poll.retries = APOLL_MAX_RETRY;
662 }
663 apoll->double_poll = NULL;
664 req->apoll = apoll;
665 if (unlikely(!--apoll->poll.retries))
666 return NULL;
667 return apoll;
668 }
669
io_arm_poll_handler(struct io_kiocb * req,unsigned issue_flags)670 int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
671 {
672 const struct io_issue_def *def = &io_issue_defs[req->opcode];
673 struct async_poll *apoll;
674 struct io_poll_table ipt;
675 __poll_t mask = POLLPRI | POLLERR | EPOLLET;
676 int ret;
677
678 if (!def->pollin && !def->pollout)
679 return IO_APOLL_ABORTED;
680 if (!io_file_can_poll(req))
681 return IO_APOLL_ABORTED;
682 if (!(req->flags & REQ_F_APOLL_MULTISHOT))
683 mask |= EPOLLONESHOT;
684
685 if (def->pollin) {
686 mask |= EPOLLIN | EPOLLRDNORM;
687
688 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
689 if (req->flags & REQ_F_CLEAR_POLLIN)
690 mask &= ~EPOLLIN;
691 } else {
692 mask |= EPOLLOUT | EPOLLWRNORM;
693 }
694 if (def->poll_exclusive)
695 mask |= EPOLLEXCLUSIVE;
696
697 apoll = io_req_alloc_apoll(req, issue_flags);
698 if (!apoll)
699 return IO_APOLL_ABORTED;
700 req->flags &= ~(REQ_F_SINGLE_POLL | REQ_F_DOUBLE_POLL);
701 req->flags |= REQ_F_POLLED;
702 ipt.pt._qproc = io_async_queue_proc;
703
704 io_kbuf_recycle(req, issue_flags);
705
706 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask, issue_flags);
707 if (ret)
708 return ret > 0 ? IO_APOLL_READY : IO_APOLL_ABORTED;
709 trace_io_uring_poll_arm(req, mask, apoll->poll.events);
710 return IO_APOLL_OK;
711 }
712
713 /*
714 * Returns true if we found and killed one or more poll requests
715 */
io_poll_remove_all(struct io_ring_ctx * ctx,struct io_uring_task * tctx,bool cancel_all)716 __cold bool io_poll_remove_all(struct io_ring_ctx *ctx, struct io_uring_task *tctx,
717 bool cancel_all)
718 {
719 unsigned nr_buckets = 1U << ctx->cancel_table.hash_bits;
720 struct hlist_node *tmp;
721 struct io_kiocb *req;
722 bool found = false;
723 int i;
724
725 lockdep_assert_held(&ctx->uring_lock);
726
727 for (i = 0; i < nr_buckets; i++) {
728 struct io_hash_bucket *hb = &ctx->cancel_table.hbs[i];
729
730 hlist_for_each_entry_safe(req, tmp, &hb->list, hash_node) {
731 if (io_match_task_safe(req, tctx, cancel_all)) {
732 hlist_del_init(&req->hash_node);
733 io_poll_cancel_req(req);
734 found = true;
735 }
736 }
737 }
738 return found;
739 }
740
io_poll_find(struct io_ring_ctx * ctx,bool poll_only,struct io_cancel_data * cd)741 static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, bool poll_only,
742 struct io_cancel_data *cd)
743 {
744 struct io_kiocb *req;
745 u32 index = hash_long(cd->data, ctx->cancel_table.hash_bits);
746 struct io_hash_bucket *hb = &ctx->cancel_table.hbs[index];
747
748 hlist_for_each_entry(req, &hb->list, hash_node) {
749 if (cd->data != req->cqe.user_data)
750 continue;
751 if (poll_only && req->opcode != IORING_OP_POLL_ADD)
752 continue;
753 if (cd->flags & IORING_ASYNC_CANCEL_ALL) {
754 if (io_cancel_match_sequence(req, cd->seq))
755 continue;
756 }
757 return req;
758 }
759 return NULL;
760 }
761
io_poll_file_find(struct io_ring_ctx * ctx,struct io_cancel_data * cd)762 static struct io_kiocb *io_poll_file_find(struct io_ring_ctx *ctx,
763 struct io_cancel_data *cd)
764 {
765 unsigned nr_buckets = 1U << ctx->cancel_table.hash_bits;
766 struct io_kiocb *req;
767 int i;
768
769 for (i = 0; i < nr_buckets; i++) {
770 struct io_hash_bucket *hb = &ctx->cancel_table.hbs[i];
771
772 hlist_for_each_entry(req, &hb->list, hash_node) {
773 if (io_cancel_req_match(req, cd))
774 return req;
775 }
776 }
777 return NULL;
778 }
779
io_poll_disarm(struct io_kiocb * req)780 static int io_poll_disarm(struct io_kiocb *req)
781 {
782 if (!req)
783 return -ENOENT;
784 if (!io_poll_get_ownership(req))
785 return -EALREADY;
786 io_poll_remove_entries(req);
787 hash_del(&req->hash_node);
788 return 0;
789 }
790
__io_poll_cancel(struct io_ring_ctx * ctx,struct io_cancel_data * cd)791 static int __io_poll_cancel(struct io_ring_ctx *ctx, struct io_cancel_data *cd)
792 {
793 struct io_kiocb *req;
794
795 if (cd->flags & (IORING_ASYNC_CANCEL_FD | IORING_ASYNC_CANCEL_OP |
796 IORING_ASYNC_CANCEL_ANY))
797 req = io_poll_file_find(ctx, cd);
798 else
799 req = io_poll_find(ctx, false, cd);
800
801 if (req) {
802 io_poll_cancel_req(req);
803 return 0;
804 }
805 return -ENOENT;
806 }
807
io_poll_cancel(struct io_ring_ctx * ctx,struct io_cancel_data * cd,unsigned issue_flags)808 int io_poll_cancel(struct io_ring_ctx *ctx, struct io_cancel_data *cd,
809 unsigned issue_flags)
810 {
811 int ret;
812
813 io_ring_submit_lock(ctx, issue_flags);
814 ret = __io_poll_cancel(ctx, cd);
815 io_ring_submit_unlock(ctx, issue_flags);
816 return ret;
817 }
818
io_poll_parse_events(const struct io_uring_sqe * sqe,unsigned int flags)819 static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe,
820 unsigned int flags)
821 {
822 u32 events;
823
824 events = READ_ONCE(sqe->poll32_events);
825 #ifdef __BIG_ENDIAN
826 events = swahw32(events);
827 #endif
828 if (!(flags & IORING_POLL_ADD_MULTI))
829 events |= EPOLLONESHOT;
830 if (!(flags & IORING_POLL_ADD_LEVEL))
831 events |= EPOLLET;
832 return demangle_poll(events) |
833 (events & (EPOLLEXCLUSIVE|EPOLLONESHOT|EPOLLET));
834 }
835
io_poll_remove_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)836 int io_poll_remove_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
837 {
838 struct io_poll_update *upd = io_kiocb_to_cmd(req, struct io_poll_update);
839 u32 flags;
840
841 if (sqe->buf_index || sqe->splice_fd_in)
842 return -EINVAL;
843 flags = READ_ONCE(sqe->len);
844 if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA |
845 IORING_POLL_ADD_MULTI))
846 return -EINVAL;
847 /* meaningless without update */
848 if (flags == IORING_POLL_ADD_MULTI)
849 return -EINVAL;
850
851 upd->old_user_data = READ_ONCE(sqe->addr);
852 upd->update_events = flags & IORING_POLL_UPDATE_EVENTS;
853 upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA;
854
855 upd->new_user_data = READ_ONCE(sqe->off);
856 if (!upd->update_user_data && upd->new_user_data)
857 return -EINVAL;
858 if (upd->update_events)
859 upd->events = io_poll_parse_events(sqe, flags);
860 else if (sqe->poll32_events)
861 return -EINVAL;
862
863 return 0;
864 }
865
io_poll_add_prep(struct io_kiocb * req,const struct io_uring_sqe * sqe)866 int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
867 {
868 struct io_poll *poll = io_kiocb_to_cmd(req, struct io_poll);
869 u32 flags;
870
871 if (sqe->buf_index || sqe->off || sqe->addr)
872 return -EINVAL;
873 flags = READ_ONCE(sqe->len);
874 if (flags & ~IORING_POLL_ADD_MULTI)
875 return -EINVAL;
876 if ((flags & IORING_POLL_ADD_MULTI) && (req->flags & REQ_F_CQE_SKIP))
877 return -EINVAL;
878
879 poll->events = io_poll_parse_events(sqe, flags);
880 return 0;
881 }
882
io_poll_add(struct io_kiocb * req,unsigned int issue_flags)883 int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
884 {
885 struct io_poll *poll = io_kiocb_to_cmd(req, struct io_poll);
886 struct io_poll_table ipt;
887 int ret;
888
889 ipt.pt._qproc = io_poll_queue_proc;
890
891 ret = __io_arm_poll_handler(req, poll, &ipt, poll->events, issue_flags);
892 if (ret > 0) {
893 io_req_set_res(req, ipt.result_mask, 0);
894 return IOU_OK;
895 }
896 return ret ?: IOU_ISSUE_SKIP_COMPLETE;
897 }
898
io_poll_remove(struct io_kiocb * req,unsigned int issue_flags)899 int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
900 {
901 struct io_poll_update *poll_update = io_kiocb_to_cmd(req, struct io_poll_update);
902 struct io_ring_ctx *ctx = req->ctx;
903 struct io_cancel_data cd = { .ctx = ctx, .data = poll_update->old_user_data, };
904 struct io_kiocb *preq;
905 int ret2, ret = 0;
906
907 io_ring_submit_lock(ctx, issue_flags);
908 preq = io_poll_find(ctx, true, &cd);
909 ret2 = io_poll_disarm(preq);
910 if (ret2) {
911 ret = ret2;
912 goto out;
913 }
914 if (WARN_ON_ONCE(preq->opcode != IORING_OP_POLL_ADD)) {
915 ret = -EFAULT;
916 goto out;
917 }
918
919 if (poll_update->update_events || poll_update->update_user_data) {
920 /* only mask one event flags, keep behavior flags */
921 if (poll_update->update_events) {
922 struct io_poll *poll = io_kiocb_to_cmd(preq, struct io_poll);
923
924 poll->events &= ~0xffff;
925 poll->events |= poll_update->events & 0xffff;
926 poll->events |= IO_POLL_UNMASK;
927 }
928 if (poll_update->update_user_data)
929 preq->cqe.user_data = poll_update->new_user_data;
930
931 ret2 = io_poll_add(preq, issue_flags & ~IO_URING_F_UNLOCKED);
932 /* successfully updated, don't complete poll request */
933 if (!ret2 || ret2 == -EIOCBQUEUED)
934 goto out;
935 }
936
937 req_set_fail(preq);
938 io_req_set_res(preq, -ECANCELED, 0);
939 preq->io_task_work.func = io_req_task_complete;
940 io_req_task_work_add(preq);
941 out:
942 io_ring_submit_unlock(ctx, issue_flags);
943 if (ret < 0) {
944 req_set_fail(req);
945 return ret;
946 }
947 /* complete update request, we're done with it */
948 io_req_set_res(req, ret, 0);
949 return IOU_OK;
950 }
951